Path: utzoo!yunexus!ists!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucsd!brian
From: brian@ucsd.Edu (Brian Kantor)
Newsgroups: comp.graphics
Subject: Re: Workstations that can record/play realtime video
Summary: NTSC = Never Twice the Same Color
Message-ID: <10135@ucsd.Edu>
Date: 18 Nov 89 18:26:09 GMT
Article-I.D.: ucsd.10135
References: <1593@ndmath.UUCP> <11390024@hpldola.HP.COM>
Reply-To: brian@ucsd.edu (Brian Kantor)
Organization: What's left of the UCSD Computer Graphics Lab
Lines: 49

Disclaimer: I'm working from what's left of my memory, so maybe some
of these numbers are off a little.

Recall that the broadcast NTSC luminance bandwidth is about 4.2 MHz,
I is 1.5 MHz, and Q is .5 MHz, so it's not necessary to store lots of
color information if you're dealing with NTSC in and out.

One older special-effects real-time video toy that I've worked on
stored NTSC as YIQ components: 492 vertical lines with 512 pixels per
line for luminance (Y), 256 for the Inphase chroma signal, and 64 for
the Quadrature chroma.  Each was 8 bits.

A back-of-the-envelope calculation (divide the bandwidth by the
horizontal scan frequency) will give you 267 pixels per line luminance,
95 I, and 31 Q: a rough indication that these pixel numbers are
probably about right - you lose just a bit of luminance, and you're
actually storing the chroma at slightly more resolution than needed.
(Don't forget Dr. Nyquist: sample at twice the max frequency.)

Just for reference, so you don't have to go look it up:

	R =   .94I +  .62Q + Y
	G =  -.27I +  .65Q + Y
	B = -1.11I - 1.70Q + Y

and
	Y = .30R + .59G + .11B
	I = .60R - .28G - .32B
	Q = .21R - .52G + .31B

Note that to save money, most television receivers do NOT directly
demodulate I and Q; instead a 33 degree shift of the chroma carrier
phase is used to demodulate R-Y, G-Y, and B-Y, with a reasonably good
approximation of the original RGB camera signal resulting.

Also, these bandwidths means that having more than 30 or so distinctly
different shades of color on an NTSC scan line isn't generally
possible; the Q carrier can't shift that fast!  An implication of this
is that one should NOT expect horizontal color wedges to look anywhere
as good as vertical ones - a fact which is good to keep in mind when
planning graphics images for broadcast use.  Aliasing in color looks
particularly bad.

I think it would be wise for any graphics house to routinely encode
their workstation images into NTSC and view them after perhaps two
generations of 1" video recording, just to make sure that the consumer
is going to see what the artist drew.  What looks really great on the
workstation monitor can really look like garbage in the home.
	- Brian